Hector Martin

A small s**tstorm is heading your way, if you're in the business of running code on Intel computers. Here's a sad story about the state of computing in 2019. It'll take a couple of tweets, but it's actually kind of important (?), so please retweet so this gets proper attention:

Is nobody attacking AMD, or is there some difference in design philosophy that has spared them from this wave of chip attacks?

This isn't a chip attack, it's a stupid design bug. That said, there *is* a difference in design philosophy. Intel has chosen performance over any kind of design sanity / safety, every time, it seems as a systematic rule. AMD hasn't.

So while both AMD and Intel have been equally hit by the general idea of speculation-based leaks in userland code (nothing the CPU can do about that), Intel has also had a huge pile of massively privilege-crossing speculation leaks that AMD hasn't.

Because the Intel philosophy seems to have been "ignore security/sanity until the instruction retires", so all kinds of batshit insane stuff happens in speculation on Intel CPUs (like bypassing guest VM page tables or loading bad or privileged data) which doesn't on AMD.

This “ignore security/sanity” sentiment that intel “has” is completely false; Intel internally has never had that philosophy. That’s something cooked up by reddit trolls.

The evidence is damning. All those speculation bugs Intel had due to cross-privilege leakage prior to retirement make it clear. It's not once or twice, it's a pattern.

By the way, I said "ignore security/sanity *until the instruction retires*". They cared about the security of the end result, they just didn't care about security along the way, and as anyone well versed in security knows, that's a recipe for disaster. Defense in depth.